rebase

Created using spr 1.3.4

flang/include/flang/Optimizer/Transforms/CufOpConversion.h → flang/include/flang/Optimizer/Transforms/CUFOpConversion.h

@@ -1,4 +1,4 @@
-//===------- Optimizer/Transforms/CufOpConversion.h -------------*- C++ -*-===//
+//===------- Optimizer/Transforms/CUFOpConversion.h -------------*- C++ -*-===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
@@ -19,7 +19,7 @@ class LLVMTypeConverter;
 namespace mlir {
 class DataLayout;
 class SymbolTable;
-}
+} // namespace mlir
 
 namespace cuf {
 

flang/include/flang/Optimizer/Transforms/Passes.h

@@ -40,7 +40,7 @@ namespace fir {
 #define GEN_PASS_DECL_CHARACTERCONVERSION
 #define GEN_PASS_DECL_CFGCONVERSION
 #define GEN_PASS_DECL_CUFADDCONSTRUCTOR
-#define GEN_PASS_DECL_CUFIMPLICITDEVICEGLOBAL
+#define GEN_PASS_DECL_CUFDEVICEGLOBAL
 #define GEN_PASS_DECL_CUFOPCONVERSION
 #define GEN_PASS_DECL_EXTERNALNAMECONVERSION
 #define GEN_PASS_DECL_MEMREFDATAFLOWOPT

flang/include/flang/Optimizer/Transforms/Passes.td

@@ -421,15 +421,15 @@ def AssumedRankOpConversion : Pass<"fir-assumed-rank-op", "mlir::ModuleOp"> {
   ];
 }
 
-def CufOpConversion : Pass<"cuf-convert", "mlir::ModuleOp"> {
+def CUFOpConversion : Pass<"cuf-convert", "mlir::ModuleOp"> {
   let summary = "Convert some CUF operations to runtime calls";
   let dependentDialects = [
     "fir::FIROpsDialect"
   ];
 }
 
-def CufImplicitDeviceGlobal :
-    Pass<"cuf-implicit-device-global", "mlir::ModuleOp"> {
+def CUFDeviceGlobal :
+    Pass<"cuf-device-global", "mlir::ModuleOp"> {
   let summary = "Flag globals used in device function with data attribute";
   let dependentDialects = [
     "cuf::CUFDialect"

flang/lib/Optimizer/Transforms/CMakeLists.txt

@@ -10,8 +10,8 @@ add_flang_library(FIRTransforms
   ConstantArgumentGlobalisation.cpp
   ControlFlowConverter.cpp
   CUFAddConstructor.cpp
-  CufImplicitDeviceGlobal.cpp
-  CufOpConversion.cpp
+  CUFDeviceGlobal.cpp
+  CUFOpConversion.cpp
   ArrayValueCopy.cpp
   ExternalNameConversion.cpp
   MemoryUtils.cpp

flang/lib/Optimizer/Transforms/CufImplicitDeviceGlobal.cpp → flang/lib/Optimizer/Transforms/CUFDeviceGlobal.cpp

@@ -1,4 +1,4 @@
-//===-- CufOpConversion.cpp -----------------------------------------------===//
+//===-- CUFOpConversion.cpp -----------------------------------------------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
@@ -18,7 +18,7 @@
 #include "mlir/Transforms/DialectConversion.h"
 
 namespace fir {
-#define GEN_PASS_DEF_CUFIMPLICITDEVICEGLOBAL
+#define GEN_PASS_DEF_CUFDEVICEGLOBAL
 #include "flang/Optimizer/Transforms/Passes.h.inc"
 } // namespace fir
 
@@ -45,8 +45,7 @@ static void prepareImplicitDeviceGlobals(mlir::func::FuncOp funcOp,
   }
 }
 
-class CufImplicitDeviceGlobal
-    : public fir::impl::CufImplicitDeviceGlobalBase<CufImplicitDeviceGlobal> {
+class CUFDeviceGlobal : public fir::impl::CUFDeviceGlobalBase<CUFDeviceGlobal> {
 public:
   void runOnOperation() override {
     mlir::Operation *op = getOperation();

flang/lib/Optimizer/Transforms/CufOpConversion.cpp → flang/lib/Optimizer/Transforms/CUFOpConversion.cpp

@@ -1,12 +1,12 @@
-//===-- CufOpConversion.cpp -----------------------------------------------===//
+//===-- CUFDeviceGlobal.cpp -----------------------------------------------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 
-#include "flang/Optimizer/Transforms/CufOpConversion.h"
+#include "flang/Optimizer/Transforms/CUFOpConversion.h"
 #include "flang/Common/Fortran.h"
 #include "flang/Optimizer/Builder/Runtime/RTBuilder.h"
 #include "flang/Optimizer/CodeGen/TypeConverter.h"
@@ -619,7 +619,7 @@ struct CufDataTransferOpConversion
   const mlir::SymbolTable &symtab;
 };
 
-class CufOpConversion : public fir::impl::CufOpConversionBase<CufOpConversion> {
+class CUFOpConversion : public fir::impl::CUFOpConversionBase<CUFOpConversion> {
 public:
   void runOnOperation() override {
     auto *ctx = &getContext();

flang/test/Fir/CUDA/cuda-implicit-device-global.f90

@@ -1,4 +1,4 @@
-// RUN: fir-opt --split-input-file --cuf-implicit-device-global %s | FileCheck %s
+// RUN: fir-opt --split-input-file --cuf-device-global %s | FileCheck %s
 
 // Test that global used in device function are flagged with the correct
 // attribute.

llvm/lib/CodeGen/MachineOperand.cpp

@@ -1244,7 +1244,8 @@ void MachineMemOperand::print(raw_ostream &OS, ModuleSlotTracker &MST,
   }
   MachineOperand::printOperandOffset(OS, getOffset());
   if (!getSize().hasValue() ||
-      getAlign() != getSize().getValue().getKnownMinValue())
+      (!getSize().isZero() &&
+       getAlign() != getSize().getValue().getKnownMinValue()))
     OS << ", align " << getAlign().value();
   if (getAlign() != getBaseAlign())
     OS << ", basealign " << getBaseAlign().value();

llvm/lib/Target/NVPTX/MCTargetDesc/NVPTXTargetStreamer.cpp

@@ -11,9 +11,13 @@
 //===----------------------------------------------------------------------===//
 
 #include "NVPTXTargetStreamer.h"
+#include "NVPTXUtilities.h"
 #include "llvm/MC/MCAsmInfo.h"
 #include "llvm/MC/MCContext.h"
+#include "llvm/MC/MCExpr.h"
 #include "llvm/MC/MCObjectFileInfo.h"
+#include "llvm/MC/MCSymbol.h"
+#include "llvm/Support/Casting.h"
 
 using namespace llvm;
 
@@ -135,3 +139,16 @@ void NVPTXTargetStreamer::emitRawBytes(StringRef Data) {
 #endif
 }
 
+void NVPTXTargetStreamer::emitValue(const MCExpr *Value) {
+  if (Value->getKind() == MCExpr::SymbolRef) {
+    const MCSymbolRefExpr &SRE = cast<MCSymbolRefExpr>(*Value);
+    StringRef SymName = SRE.getSymbol().getName();
+    if (!SymName.starts_with(".debug")) {
+      Streamer.emitRawText(NVPTX::getValidPTXIdentifier(SymName));
+      return;
+    }
+    // Fall through to the normal printing.
+  }
+  // Otherwise, print the Value normally.
+  MCTargetStreamer::emitValue(Value);
+}

llvm/lib/Target/NVPTX/MCTargetDesc/NVPTXTargetStreamer.h

@@ -48,6 +48,8 @@ class NVPTXTargetStreamer : public MCTargetStreamer {
   ///
   /// This is used to emit bytes in \p Data as sequence of .byte directives.
   void emitRawBytes(StringRef Data) override;
+  /// Makes sure that labels are mangled the same way as the actual symbols.
+  void emitValue(const MCExpr *Value) override;
 };
 
 class NVPTXAsmTargetStreamer : public NVPTXTargetStreamer {

llvm/lib/Target/NVPTX/NVPTXAssignValidGlobalNames.cpp

@@ -17,13 +17,11 @@
 //===----------------------------------------------------------------------===//
 
 #include "NVPTX.h"
-#include "llvm/ADT/StringExtras.h"
+#include "NVPTXUtilities.h"
 #include "llvm/IR/Function.h"
 #include "llvm/IR/GlobalVariable.h"
 #include "llvm/IR/LegacyPassManager.h"
 #include "llvm/IR/Module.h"
-#include "llvm/Support/raw_ostream.h"
-#include <string>
 
 using namespace llvm;
 
@@ -35,11 +33,8 @@ class NVPTXAssignValidGlobalNames : public ModulePass {
   NVPTXAssignValidGlobalNames() : ModulePass(ID) {}
 
   bool runOnModule(Module &M) override;
-
-  /// Clean up the name to remove symbols invalid in PTX.
-  std::string cleanUpName(StringRef Name);
 };
-}
+} // namespace
 
 char NVPTXAssignValidGlobalNames::ID = 0;
 
@@ -58,34 +53,18 @@ bool NVPTXAssignValidGlobalNames::runOnModule(Module &M) {
       // Note: this does not create collisions - if setName is asked to set the
       // name to something that already exists, it adds a proper postfix to
       // avoid collisions.
-      GV.setName(cleanUpName(GV.getName()));
+      GV.setName(NVPTX::getValidPTXIdentifier(GV.getName()));
     }
   }
 
   // Do the same for local functions.
   for (Function &F : M.functions())
     if (F.hasLocalLinkage())
-      F.setName(cleanUpName(F.getName()));
+      F.setName(NVPTX::getValidPTXIdentifier(F.getName()));
 
   return true;
 }
 
-std::string NVPTXAssignValidGlobalNames::cleanUpName(StringRef Name) {
-  std::string ValidName;
-  raw_string_ostream ValidNameStream(ValidName);
-  for (char C : Name) {
-    // While PTX also allows '%' at the start of identifiers, LLVM will throw a
-    // fatal error for '%' in symbol names in MCSymbol::print. Exclude for now.
-    if (isAlnum(C) || C == '_' || C == '$') {
-      ValidNameStream << C;
-    } else {
-      ValidNameStream << "_$_";
-    }
-  }
-
-  return ValidNameStream.str();
-}
-
 ModulePass *llvm::createNVPTXAssignValidGlobalNamesPass() {
   return new NVPTXAssignValidGlobalNames();
 }

llvm/lib/Target/NVPTX/NVPTXUtilities.cpp

@@ -13,6 +13,7 @@
 #include "NVPTXUtilities.h"
 #include "NVPTX.h"
 #include "NVPTXTargetMachine.h"
+#include "llvm/ADT/StringExtras.h"
 #include "llvm/ADT/StringRef.h"
 #include "llvm/IR/Constants.h"
 #include "llvm/IR/Function.h"
@@ -33,7 +34,7 @@
 namespace llvm {
 
 namespace {
-typedef std::map<std::string, std::vector<unsigned> > key_val_pair_t;
+typedef std::map<std::string, std::vector<unsigned>> key_val_pair_t;
 typedef std::map<const GlobalValue *, key_val_pair_t> global_val_annot_t;
 
 struct AnnotationCache {

llvm/lib/Target/NVPTX/NVPTXUtilities.h

@@ -14,6 +14,7 @@
 #define LLVM_LIB_TARGET_NVPTX_NVPTXUTILITIES_H
 
 #include "NVPTX.h"
+#include "llvm/ADT/StringExtras.h"
 #include "llvm/CodeGen/ValueTypes.h"
 #include "llvm/IR/Function.h"
 #include "llvm/IR/GlobalVariable.h"
@@ -84,6 +85,19 @@ bool shouldEmitPTXNoReturn(const Value *V, const TargetMachine &TM);
 bool Isv2x16VT(EVT VT);
 
 namespace NVPTX {
+inline std::string getValidPTXIdentifier(StringRef Name) {
+  std::string ValidName;
+  ValidName.reserve(Name.size() + 4);
+  for (char C : Name)
+    // While PTX also allows '%' at the start of identifiers, LLVM will throw a
+    // fatal error for '%' in symbol names in MCSymbol::print. Exclude for now.
+    if (isAlnum(C) || C == '_' || C == '$')
+      ValidName.push_back(C);
+    else
+      ValidName.append({'_', '$', '_'});
+
+  return ValidName;
+}
 
 inline std::string OrderingToString(Ordering Order) {
   switch (Order) {

llvm/test/CodeGen/WebAssembly/externref-globalget.ll

@@ -1,4 +1,7 @@
 ; RUN: llc < %s --mtriple=wasm32-unknown-unknown -asm-verbose=false -mattr=+reference-types | FileCheck %s
+; Test for MIR printing of reference types in other address spaces. This should
+; not error out.
+; RUN: llc < %s --mtriple=wasm32-unknown-unknown -asm-verbose=false -mattr=+reference-types -print-after=finalize-isel | FileCheck %s
 
 %externref = type ptr addrspace(10) ;; addrspace 10 is nonintegral
 

llvm/test/DebugInfo/NVPTX/debug-ptx-symbols.ll

@@ -0,0 +1,42 @@
+; RUN: llc < %s -march=nvptx64 -mcpu=sm_60 | FileCheck %s
+; RUN: %if ptxas %{ llc < %s -march=nvptx64 -mcpu=sm_60 | %ptxas-verify %}
+target datalayout = "e-i64:64-i128:128-v16:16-v32:32-n16:32:64"
+target triple = "nvptx64-nvidia-cuda"
+
+; Verify that the symbols with the characters illegal in PTX get appropriately mangled.
+@__PRETTY_FUNCTION__._Z3foov = private unnamed_addr constant [11 x i8] c"void foo()\00", align 1, !dbg !0
+; '.' gets replaced with `_$_`.
+; CHECK: .global .align 1 .b8 __PRETTY_FUNCTION___$__Z3foov[11] = {118, 111, 105, 100, 32, 102, 111, 111, 40, 41};
+
+; .debug* section names are special and are allowed to have the leading dot.
+; CHECK-DAG: .section        .debug_abbrev
+; CHECK-DAG: .section        .debug_info
+; CHECK-DAG: .b32 .debug_abbrev
+; CHECK-DAG: .b32 .debug_line
+; CHECK-DAG: .section        .debug_macinfo
+
+; .. but the symbol name must be mangled the same way here as it was at the definition point.
+; CHECK-DAG: .b64 __PRETTY_FUNCTION___$__Z3foov
+;
+
+!llvm.dbg.cu = !{!8}
+!llvm.module.flags = !{!10, !11, !12}
+
+!0 = !DIGlobalVariableExpression(var: !1, expr: !DIExpression())
+!1 = distinct !DIGlobalVariable(scope: null, file: !2, line: 1, type: !3, isLocal: true, isDefinition: true)
+!2 = !DIFile(filename: "<stdin>", directory: "/usr/local/google/home/tra/work/llvm/build/release+assert+zapcc/dbg-dot")
+!3 = !DICompositeType(tag: DW_TAG_array_type, baseType: !4, size: 88, elements: !6)
+!4 = !DIDerivedType(tag: DW_TAG_const_type, baseType: !5)
+!5 = !DIBasicType(name: "char", size: 8, encoding: DW_ATE_signed_char)
+!6 = !{!7}
+!7 = !DISubrange(count: 11)
+!8 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus_14, file: !2, producer: "clang version 20.0.0git", isOptimized: false, runtimeVersion: 0, emissionKind: FullDebug, globals: !9, splitDebugInlining: false, nameTableKind: None)
+!9 = !{!0}
+!10 = !{i32 2, !"Debug Info Version", i32 3}
+!11 = !{i32 1, !"wchar_size", i32 4}
+!12 = !{i32 4, !"nvvm-reflect-ftz", i32 0}
+!13 = !{!"clang version 20.0.0git"}
+!14 = distinct !DISubprogram(name: "foo", linkageName: "_Z3foov", scope: !2, file: !2, line: 1, type: !15, scopeLine: 1, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition, unit: !8)
+!15 = !DISubroutineType(types: !16)
+!16 = !{null}
+!17 = !DILocation(line: 1, column: 56, scope: !14)